AEE
Agent Execution Partnership · Interactive Demo

The Action Execution Engine — every agent action authorized, observable, and verifiable.

Propose an action from an agent, bind it to a task contract, and watch it run through the closed loop: policy evaluation, approval gates, bounded execution, and post-condition verification. Everything is captured in an immutable audit ledger.

Actorc9z8ch3t
Can approve / deny at the human-in-the-loop gate.
03
Closed Loop
Observe → Propose → Policy → Approve → Execute → Verify → Decide
Risk classCONSEQUENTIAL_WRITE
Stage 01Idle
Task Contract
Goal, constraints, risk budget
Stage 02Idle
Observe State
Snapshot pre-conditions
Stage 03Idle
Propose Action
One bounded action
Stage 04Idle
Policy Evaluate
Identity, authority, risk
Stage 05Idle
Approval Gate
Human-in-the-loop if required
Stage 06Idle
Execute
One channel, bounded scope
Stage 07Idle
Verify Effects
Compare observed vs. expected
Stage 08Idle
Decide
Retain / compensate / escalate
Action request · train.run
bounded / single-channel
{
  "op": "train.run",
  "channel": "gpu_training",
  "risk_class": "CONSEQUENTIAL_WRITE",
  "targets": [
    "cluster://gpu-node-4"
  ],
  "expected_effect": "Emit candidate checkpoint; retain if BPB improves ≥0.001",
  "policy_hints": [
    "depth ≤ 12",
    "memory ≤ 40GB",
    "time ≤ 300s"
  ]
}
Idle. Choose an action and dispatch the loop — the ledger will log every stage boundary.
05
Scenario Recorder
Every completed run is captured; replay step-by-step from the ledger
No scenarios yet. Dispatch a loop and it will be recorded here for replay.